安装全称参考https://keras-cn.readthedocs.io/en/latest/for_beginners/keras_linux/
环境中已配置cuda8.0.cudnn5.0,ubuntu16.04
安装的tensorflow-gpu版本
安装完成后进行测试,import tensorflow
问题:ImportError:libcublas.so.9.0: cannot open shared object file: No such file or directory
原因:tensorflow版本与cudnn以及cuda版本不对应,参考:https://blog.csdn.net/twt520ly/article/details/79415787
解决方案,换1.3,仍然出现问题
问题: libcudnn.so.6:cannot open sharedobjectfile: No such file or directory
原因:1.3与cudnn5不兼容,换1.2 解决
新问题:import keras报错 Using TensorFlow backend. Couldn't import dot_parser, loading of dot files w
解决方式:参考https://blog.csdn.net/liyaohhh/article/details/51166268
sudo pip uninstall pyparsingsudo pip install -Iv https://pypi.python.org/packages/source/p/pyparsing/pyparsing-1.5.7.tar.gz#md5=9be0fcdcc595199c646ab317c1d9a709sudo pip install pydot
注意加sudo,否则访问被拒绝
最后,下载keras安装包,运行python mnist_mlp.py,报错
问题: TypeError: softmax() got an unexpected keyword argument 'axis'
原因:版本问题,退回keras2.1,参考https://blog.csdn.net/w5688414/article/details/80145894